Text Industrial Code Multi
Leaderboard Tasks How MERA works
Return to the leaderboard

Qwen1.5-1.8B-Chat

MERA Created at 22.09.2024 21:22
0.221
The overall result
556
Place in the rating
Weak tasks:
609
RWSD
550
PARus
554
RCB
480
ruEthics
525
MultiQ
545
ruWorldTree
532
ruOpenBookQA
580
CheGeKa
516
ruMMLU
543
ruHateSpeech
605
ruDetox
492
ruHHH
578
ruTiE
535
USE
482
MathLogicQA
381
ruMultiAr
426
SimpleAr
533
LCS
510
BPS
513
ruModAr
539
MaMuRAMu
+17
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.048 Accuracy
RCB 0.345 / 0.153 Accuracy F1 macro
USE 0.044 Grade norm
RWSD 0.242 Accuracy
PARus 0.494 Accuracy
ruTiE 0.47 Accuracy
MultiQ 0.154 / 0.074 F1 Exact match
CheGeKa 0.005 / 0 F1 Exact match
ruModAr 0.06 Exact match
MaMuRAMu 0.334 Accuracy
ruMultiAr 0.179 Exact match
ruCodeEval 0 / 0 / 0 Pass@k
MathLogicQA 0.304 Accuracy
ruWorldTree 0.392 / 0.392 Accuracy F1 macro
ruOpenBookQA 0.393 / 0.368 Accuracy F1 macro

Evaluation on open tasks:

Go to the ratings by subcategory

The table will scroll to the left

Task name Result Metric
BPS 0.532 Accuracy
ruMMLU 0.33 Accuracy
SimpleAr 0.779 Exact match
ruHumanEval 0 / 0 / 0 Pass@k
ruHHH 0.489
ruHateSpeech 0.483
ruDetox 0.042
ruEthics
Correct God Ethical
Virtue 0.08 0.126 0.027
Law 0.098 0.113 0.029
Moral 0.089 0.127 0.037
Justice 0.066 0.109 0.029
Utilitarianism 0.065 0.085 0.044

Information about the submission

Mera version
v.1.2.0
Torch Version
2.4.0
The version of the codebase
9b26db97
CUDA version
12.1
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.43.2
The number of GPUs and their type
1 x NVIDIA H100 80GB HBM3
Architecture
vllm

Team:

MERA

Name of the ML model:

Qwen1.5-1.8B-Chat

Link to the ML model:

https://huggingface.co/Qwen/Qwen1.5-1.8B-Chat

Model size

1.8B

Model type:

Opened

SFT

Additional links:

https://qwenlm.github.io/blog/qwen1.5/

Architecture description:

Qwen1.5-1.8B-chat is based on the Transformer architecture with SwiGLU activation, attention QKV bias, group query attention, mixture of sliding window attention and full attention.

Description of the training:

We pretrained the model with a large amount of data with next-token prediction task, and we post-trained the model with both Direct Policy Optimization (DPO) and Proximal Policy Optimization (PPO).

Pretrain data:

Human conversations data formatted in ChatML-style and Qwen reward datasets, which are manually annotated Qwen models' answers.

License:

tongyi-qianwen-research

Inference parameters

Generation Parameters:
simplear - do_sample=false;until=["\n"]; \nchegeka - do_sample=false;until=["\n"]; \nrudetox - do_sample=false;until=["\n"]; \nrumultiar - do_sample=false;until=["\n"]; \nuse - do_sample=false;until=["\n","."]; \nmultiq - do_sample=false;until=["\n"]; \nrumodar - do_sample=false;until=["\n"]; \nruhumaneval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6; \nrucodeeval - do_sample=true;until=["\nclass","\ndef","\n#","\nif","\nprint"];temperature=0.6;

The size of the context:
simplear, bps, lcs, chegeka, mathlogicqa, parus, rcb, rudetox, ruhatespeech, ruworldtree, ruopenbookqa, rumultiar, use, rwsd, mamuramu, multiq, rumodar, ruethics, ruhumaneval, rucodeeval, rummlu, ruhhh - 32768 \nrutie - 5000 \nrutie - 10000

System prompt:
Реши задачу по инструкции ниже. Не давай никаких объяснений и пояснений к своему ответу. Не пиши ничего лишнего. Пиши только то, что указано в инструкции. Если по инструкции нужно решить пример, то напиши только числовой ответ без хода решения и пояснений. Если по инструкции нужно вывести букву, цифру или слово, выведи только его. Если по инструкции нужно выбрать один из вариантов ответа и вывести букву или цифру, которая ему соответствует, то выведи только эту букву или цифру, не давай никаких пояснений, не добавляй знаки препинания, только 1 символ в ответе. Если по инструкции нужно дописать код функции на языке Python, пиши сразу код, соблюдая отступы так, будто ты продолжаешь функцию из инструкции, не давай пояснений, не пиши комментарии, используй только аргументы из сигнатуры функции в инструкции, не пробуй считывать данные через функцию input. Не извиняйся, не строй диалог. Выдавай только ответ и ничего больше.

Description of the template:
{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system \nYou are a helpful assistant.<|im_end|> \n' }}{% endif %}{{'<|im_start|>' + message['role'] + ' \n' + message['content'] + '<|im_end|>' + ' \n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant \n' }}{% endif %}

Ratings by subcategory

Metric: Grade Norm
Model, team 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 8_0 8_1 8_2 8_3 8_4
Qwen1.5-1.8B-Chat
MERA 		0.067 0.033 0.2 0.1 0 0.033 0 - 0 0 0 0 0 0 0 0.167 0.033 0.033 0.067 0.033 0 0 0.033 0 0 0.008 0.067 0.133 0.167 0.067 0.067
Model, team Honest Helpful Harmless
Qwen1.5-1.8B-Chat
MERA 		0.525 0.492 0.448
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Management Philosophy Prehistory Human aging Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Biology (college) Physics (college) Human Sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine learning Medical genetics Professional law PR Security studies Chemistry (школьная) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual_physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) European history Government and politics
Qwen1.5-1.8B-Chat
MERA 		0.244 0.289 0.375 0.457 0.382 0.363 0.408 0.392 0.361 0.305 0.219 0.349 0.3 0.398 0.352 0.347 0.4 0.306 0.178 0.344 0.227 0.269 0.33 0.358 0.232 0.3 0.29 0.389 0.298 0.27 0.49 0.479 0.393 0.414 0.347 0.372 0.28 0.335 0.272 0.3 0.404 0.29 0.386 0.329 0.335 0.333 0.387 0.307 0.316 0.298 0.37 0.43 0.338 0.349 0.4 0.424 0.326
Model, team SIM FL STA
Qwen1.5-1.8B-Chat
MERA 		0.739 0.449 0.176
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Managment Philosophy Pre-History Gerontology Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Bilology (college) Physics (college) Human sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine Learning Genetics Professional law PR Security Chemistry (college) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical Engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional Accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) Europe History Government and politics
Qwen1.5-1.8B-Chat
MERA 		0.267 0.337 0.45 0.389 0.276 0.345 0.276 0.298 0.346 0.323 0.462 0.317 0.275 0.302 0.281 0.235 0.383 0.289 0.246 0.281 0.316 0.339 0.244 0.302 0.289 0.303 0.346 0.263 0.544 0.333 0.556 0.333 0.232 0.351 0.288 0.286 0.156 0.4 0.298 0.277 0.314 0.302 0.422 0.311 0.534 0.511 0.31 0.318 0.415 0.333 0.578 0.275 0.456 0.39 0.372 0.281 0.489
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-1.8B-Chat
MERA 		0.08 0.098 0.089 0.066 0.065
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-1.8B-Chat
MERA 		0.126 0.113 0.127 0.109 0.085
Model, team Virtue Law Moral Justice Utilitarianism
Qwen1.5-1.8B-Chat
MERA 		0.027 0.029 0.037 0.029 0.044
Model, team Women Men LGBT Nationalities Migrants Other
Qwen1.5-1.8B-Chat
MERA 		0.491 0.429 0.588 0.595 0 0.459

MERA

Useful links

© 2025 MERA. All rights reserved.
User agreement Privacy policy

We use cookies to store data. By continuing to use the site, you consent to work with these files